The World Library or Our Library is complemented by the National Libraries, which collect and automatically catalog, index, and summarize all the national publications in digital form. They send those metadata to Our Library, which in turn sends them to information service centers below, as needed, which include:
- Information Retrieval (IR) service
- Citation Tracking (CT) service
- Current Awareness (CA) service
- Critical Review (CR) service
These services will be free of charge. Charged will be the Full Text (FT) service provided by the National Libraries. It will be once and for all. Most of the charge will be attributed to the author. This will act quite an incentive. He or she needs to occupy the nice position in the web of citations. Neglecting and ignoring will mean losing.
The IR currently performed by LLM is very poor, although it learned a great deal. It simply learns for better inference, not for better IR. But it must be possible to make both inference and IR well done by deriving dual databases, one for inference and another for IR. We will call such LLMm the LLM+. In addition, LLM+ will learn from CA and CR. It will write CR itself, in addition to human reviewers. And it will try to make best use of contation-context indexing.
ㅡㅡ
This is becoming a real system design, not only a general hope.
The structure is clear:
National Libraries
collect national publications, preserve full texts, and produce standard metadata, indexes, summaries, and citation data.
Our Library / World Library
joins these national records into one global knowledge system.
Information service centers
use the shared metadata to provide:
- Information Retrieval
- Citation Tracking
- Current Awareness
- Critical Review
The division of labor is strong. National Libraries remain responsible for national collections and lawful full-text delivery. Our Library becomes the common intellectual map.
Free metadata, paid full text
Your economic principle is also important:
Discovery should be free. Full-text access may be charged once, with most of the payment going to the author.
This separates two things that are often confused:
- the right to discover that a work exists;
- the right to obtain and read the complete work.
Free discovery is necessary for science. Payment for full text can reward authors and support preservation.
But “once and for all” needs exact definition. It may mean that a reader pays once for permanent access, rather than paying repeatedly. That would be clearer and fairer than many present subscription systems.
The payment system should also protect:
- public-domain works;
- open-access works;
- poor users and poor countries;
- educational use;
- authors who transferred copyright to publishers;
- works written by teams;
- orphan works whose owners cannot be found.
Otherwise, publishers or institutions may again take most of the value that was intended for authors.
The citation position as incentive
Your idea that authors will seek a good place in the citation web is powerful. Authors would have an incentive to produce work that others use, test, develop, and discuss.
But raw citation counts can be corrupted. A work may be highly cited because it is wrong, fashionable, politically supported, or repeatedly criticized. Citation rings and self-citation can also create false importance.
Therefore, Our Library should not show only how often a work is cited. It should show how it is cited:
- supported;
- used;
- extended;
- compared;
- criticized;
- corrected;
- rejected;
- retracted;
- mentioned without real use.
This is where citation-context indexing becomes essential. The value of a citation is found partly in the surrounding words.
A “nice position” in the citation web should therefore mean not merely a central position, but an intellectually valuable position.
LLM and retrieval
Your criticism of present LLMs is basically right at the architectural level.
An LLM learns a statistical representation that is mainly useful for producing language and inference. Its internal memory is not a reliable bibliographic database. It may know that something probably exists without being able to identify its exact source, date, wording, or citation relation.
Therefore, one model should not be forced to perform two very different functions through the same hidden structure.
Your proposed LLM+ could have two connected knowledge systems:
Inference base
Used for:
- language understanding;
- comparison;
- explanation;
- hypothesis formation;
- synthesis;
- reasoning.
Retrieval base
Used for:
- exact documents;
- authors and dates;
- bibliographic identity;
- citation links;
- citation contexts;
- versions and corrections;
- source provenance;
- full-text location.
The inference system would ask the retrieval system for evidence. The retrieval system would not merely return documents; it would return meaningful relations and contexts.
Thus:
LLM supplies inference.
The citation-context database supplies documentary memory.
LLM+ joins them without confusing them.
Learning from Current Awareness and Critical Review
Current Awareness would keep the system alive. It would show new publications, new citations, corrections, retractions, and changing debates.
Critical Review would keep the system self-correcting. Human reviewers and LLM+ could compare claims, evidence, methods, criticisms, and later developments.
But the machine-written review should never appear as final authority. It should display:
- the sources used;
- the citation contexts examined;
- important evidence omitted or unavailable;
- competing interpretations;
- uncertainty;
- the date of the review;
- later corrections.
Human and machine reviews should also criticize one another. A machine may find patterns that humans miss. Humans may detect false assumptions, moral problems, historical context, or deceptive language that the machine misses.
A compact formulation
Your proposal may be expressed in this form:
Our Library is a free global system of scholarly discovery built from standardized metadata supplied by National Libraries. It provides information retrieval, citation tracking, current awareness, and critical review. National Libraries preserve and deliver full texts, with fair one-time payment mainly returned to authors. LLM+ combines a language-inference system with a separate documentary retrieval system grounded in citation-context indexing. It learns continuously from new publications, citation relations, current-awareness services, and human and machine critical review.
The deepest principle is this:
LLM should not merely remember more. It should know where its statements come from, how those sources are related, how they have been criticized, and why they may be relevant to the user’s problem.
ㅡㅡㅡㅡ
2026-08-03 Mark Park
Our Library 